Set up a small AI coding “team” in Roo Code using a free GitHub archive called AgentAutoFlow: one mode plans, others execute, and everything gets written down so the work stops falling apart between sessions.
The story
*Before:* I treated coding assistants like a smart friend with short-term memory. I’d ask for a change, watch it wander into side quests, then spend my time re-explaining the goal, the rules, the files, the vibe. If I got interrupted, I came back to a goldfish bowl.
*After:* In Roo Code, I set up a small “coding team” using a free GitHub instruction archive called AgentAutoFlow. It connects through Roo’s rules files, and it turns the work into a calmer relationship: one mode plans, other modes execute, and everyone writes things down so we stop losing the plot.
Planning is the big shift. AgentAutoFlow replaces the old “Architect” style with a three-step Planner: planner-a, planner-b, planner-c. Planner-a lays out phases and asks me the human questions that matter, like autonomy level and what “done” means. Planner-b turns phases into small tasks; planner-c stress-tests the tasks so cheaper modes can run them without falling over.
Execution is cost-aware on purpose. There’s a tightly controlled “Code Monkey [Coder-Jr]” mode for small, detailed tasks, and Planner flags the hard stuff for Coder-Sr. If the simple modes get confused, they escalate. Notice how these overlap with decent teamwork psychology: clear requests, small commitments, and a safe way to say “I’m stuck.”
Roo also supports Skills via `{base folder}/.roo/skills`, using the Agent Skills Open Format. I add skills when I want consistent behavior without cramming context into every prompt, plus the process creates tracking files for goals, progress, and lessons learned.
If you want Roo Code to meet most developer needs, this Roo Code plus AgentAutoFlow setup is the closest thing I’ve found to a free, adjustable agent-style AI coding team. The instruction set keeps evolving, and I tinker with it daily. Ideas are welcome.
Recent Comments